面向小目标与模糊目标的轨道缺陷检测框架

doi:10.3785/j.issn.1008-973X.2026.11.014

浙江大学学报(工学版)

2026, Vol. 60

Issue (11): 2453-2461 DOI: 10.3785/j.issn.1008-973X.2026.11.014

计算机技术

面向小目标与模糊目标的轨道缺陷检测框架

黄腾飞1(

),火久元1,2,*(

),曹芳1,武泽宇1,孔垂乐1

1. 兰州交通大学电子与信息工程学院，甘肃兰州 730070
2. 国家冰川冻土沙漠科学数据中心，甘肃兰州 730000

Framework for railway defect detection oriented to small and blurry object

Tengfei HUANG1(

),Jiuyuan HUO1,2,*(

),Fang CAO1,Zeyu WU1,Chuile KONG1

1. School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
2. National Cryosphere Desert Data Center, Lanzhou 730000, China

全文: PDF(6359 KB) HTML

摘要：

为了解决轨道缺陷检测中小目标易漏检与模糊目标易误检的问题，提出动态多尺度与纹理感知框架（DMT）. 从动态尺度感知池化（DSP）、纹理对比增强（TCE）和模糊边缘感知（BBA）出发，系统提升模型在复杂场景下的特征提取与边缘表征能力. 基于DMT框架，设计DSP、TCE与BBA 3种轻量级子模块融入检测网络. 将所提框架嵌入YOLOv8，构建名为DMT-YOLOv8的轨道缺陷检测模型. 在真实轨道缺陷数据集上的实验结果表明，DMT-YOLOv8在小目标检测与模糊目标检测方面均优于基线模型，具备良好的通用性与泛化性. DMT-YOLOv8的mAP@0.5与mAP@0.5:0.95分别达到89.1%和50.2%，整体性能超越多种主流检测方法.

关键词： 轨道缺陷检测; 小目标检测; 模糊目标; 多尺度特征融合; 纹理对比增强; 边缘模糊感知

Abstract:

To address the problems of missed detection of small objects and false detection of blurred objects in railway defect detection, a dynamic multi-scale and texture-aware framework (DMT) was proposed. Through dynamic scale-aware pooling (DSP), texture contrast enhancement (TCE), and blurred boundary awareness (BBA), the model’s feature extraction and boundary representation capabilities were enhanced in complex scenarios. Based on the DMT framework, three lightweight modules, DSP, TCE, and BBA, were designed and integrated into the detection network. The proposed framework was incorporated into YOLOv8 to construct a railway defect detection model termed DMT-YOLOv8. Experimental results on real-world railway defect datasets showed that DMT-YOLOv8 outperformed the baseline model in both small-object and blurred-object detection and exhibited good generality and generalization capability. Specifically, the mAP@0.5 and mAP@0.5:0.95 reached 89.1% and 50.2%, respectively, and the overall performance surpassed that of several mainstream detection methods.

Key words: railway defect detection small object detection blurry object multi-scale feature fusion texture contrast enhancement blurred-boundary awareness

收稿日期: 2025-09-26 出版日期: 2026-10-16

CLC:

TP 391

基金资助: 国家自然科学基金资助项目（62262038）；甘肃省科技计划-中小企业创新基金资助项目（25CXGA014）；甘肃省科技创新引导计划-科技专员专项（25CXGA030）.

通讯作者: 火久元 E-mail: 849138454@qq.com;huojy@mail.lzjtu.cn

作者简介: 黄腾飞（1996—），男，博士生，从事轨道缺陷检测研究. orcid.org/0009-0005-3588-6803. E-mail：849138454@qq.com

	服务
	把本文推荐给朋友
	加入引用管理器
	E-mail Alert
	作者相关文章
	黄腾飞
	火久元
	曹芳
	武泽宇
	孔垂乐

引用本文:

黄腾飞,火久元,曹芳,武泽宇,孔垂乐. 面向小目标与模糊目标的轨道缺陷检测框架[J]. 浙江大学学报(工学版), 2026, 60(11): 2453-2461.

Tengfei HUANG,Jiuyuan HUO,Fang CAO,Zeyu WU,Chuile KONG. Framework for railway defect detection oriented to small and blurry object. Journal of ZheJiang University (Engineering Science), 2026, 60(11): 2453-2461.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2026.11.014 或 https://www.zjujournals.com/eng/CN/Y2026/V60/I11/2453

图 1 动态尺度感知池化模块的网络结构

图 2 轻量级通道注意力机制的网络结构

图 3 纹理对比增强模块的网络结构

图 4 模糊边缘感知模块的网络结构

图 5 DMT-YOLOv8轨道缺陷检测模型的架构

表 1 动态尺度感知池化不同分组对所提模型检测性能的影响

表 2 轻量级通道注意力机制不同卷积核大小对所提模型检测性能的影响

表 3 不同模块组合对模型检测性能的影响

表 4 动态多尺度与纹理感知框架在不同基准算法上的性能对比

表 5 不同算法在轨道缺陷检测任务中的性能对比

图 6 不同检测算法PR曲线对比

图 7 不同算法在轨道缺陷数据集上的可视化检测结果

表 6 Rail-1118数据集上2种检测算法的性能对比结果

图 8 Rail-1118数据集上2种算法的可视化检测结果

图 9 所提模型检测错误样例

表 7 所提模型检测错误类型统计

1	董超群, 汪战, 廖平, 等轻量化YOLOv5s-OCG的轨枕裂纹检测算法[J]. 浙江大学学报: 工学版, 2025, 59 (9): 1838- 1845 DONG Chaoqun, WANG Zhan, LIAO Ping, et al Lightweight YOLOv5s-OCG rail sleeper crack detection algorithm[J]. Journal of Zhejiang University: Engineering Science, 2025, 59 (9): 1838- 1845
2	吴永军, 崔灿, 何永福基于语义增广与YOLOv8的钢轨表面缺陷检测方法[J]. 铁道科学与工程学报, 2024, 21 (9): 3864- 3875 WU Yongjun, CUI Can, HE Yongfu Rail surface defect detection based on semantic augmentation and YOLOv8[J]. Journal of Railway Science and Engineering, 2024, 21 (9): 3864- 3875 doi: 10.19713/j.cnki.43-1423/u.T20231930
3	GIBERT X, PATEL V M, CHELLAPPA R Deep multitask learning for railway track inspection[J]. IEEE Transactions on Intelligent Transportation Systems, 2017, 18 (1): 153- 164 doi: 10.1109/TITS.2016.2568758
4	金学松, 李霞, 李伟, 等铁路钢轨波浪形磨损研究进展[J]. 西南交通大学学报, 2016, 51 (2): 264- 273 JIN Xuesong, LI Xia, LI Wei, et al Review of rail corrugation progress[J]. Journal of Southwest Jiaotong University, 2016, 51 (2): 264- 273 doi: 10.3969/j.issn.0258-2724.2016.02.006
5	高天赐, 史一帆, 江乐鹏, 等基于瓦片编码网络的钢轨焊缝几何不平顺识别[J]. 铁道科学与工程学报, 2025, 22 (5): 2346- 2354 GAO Tianci, SHI Yifan, JIANG Lepeng, et al A tile-coding-based neural network for rail weld irregularity identification[J]. Journal of Railway Science and Engineering, 2025, 22 (5): 2346- 2354 doi: 10.19713/j.cnki.43-1423/u.T20241206
6	PHUSAKULKAJORN W, ZENG Y, LI Z, et al Unsupervised representation learning for monitoring rail infrastructures with high-frequency moving vibration sensors[J]. IEEE Transactions on Intelligent Transportation Systems, 2025, 26 (8): 12746- 12760 doi: 10.1109/TITS.2025.3557712
7	CHEN Z, WANG Q, HE Q, et al CUFuse: camera and ultrasound data fusion for rail defect detection[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23 (11): 21971- 21983 doi: 10.1109/TITS.2022.3189677
8	NI X, MA Z, LIU J, et al Attention network for rail surface defect detection via consistency of intersection-over-union(IoU)-guided center-point estimation[J]. IEEE Transactions on Industrial Informatics, 2022, 18 (3): 1694- 1705 doi: 10.1109/TII.2021.3085848
9	姚宗伟, 杨宏飞, 胡际勇, 等基于机器视觉和卷积神经网络的轨道表面缺陷检测方法[J]. 铁道学报, 2021, 43 (4): 101- 107 YAO Zongwei, YANG Hongfei, HU Jiyong, et al Track surface defect detection method based on machine vision and convolutional neural network[J]. Journal of the China Railway Society, 2021, 43 (4): 101- 107 doi: 10.3969/j.issn.1001-8360.2021.04.013
10	徐心慈, 师秀霞, 耿晨歌, 等基于深度学习的铁轨图像识别技术[J]. 铁道科学与工程学报, 2024, 21 (12): 5232- 5241 XU Xinci, SHI Xiuxia, GENG Chenge, et al Rail image recognition technology based on deep learning[J]. Journal of Railway Science and Engineering, 2024, 21 (12): 5232- 5241 doi: 10.19713/j.cnki.43-1423/u.T20240411
11	YE J, STEWART E, CHEN Q, et al Deep learning and laser-based 3-D pixel-level rail surface defect detection method[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 2513612 doi: 10.1109/tim.2023.3272033
12	刘子豪, 张佳欣, 薛峰, 等基于改进YOLO-v8的精密管件表面缺陷检测方法[J]. 浙江大学学报: 工学版, 2025, 59 (7): 1514- 1522 LIU Zihao, ZHANG Jiaxin, XUE Feng, et al Surface defect detection method of precision pipe fittings based on improved YOLO-v8[J]. Journal of Zhejiang University: Engineering Science, 2025, 59 (7): 1514- 1522
13	ZHAO Q, FANG S, LI Y, et al Track defect detection based on improved YOLOv5s[J]. IEEE Transactions on Industrial Informatics, 2025, 21 (4): 3346- 3355 doi: 10.1109/TII.2024.3523593
14	WANG B, ZHOU W, YAN W, et al PENet-KD: progressive enhancement network via knowledge distillation for rail surface defect detection[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 5032811 doi: 10.1109/tim.2023.3330222
15	ZHANG Y, LIU C Real-time pavement damage detection with damage shape adaptation[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25 (11): 18954- 18963 doi: 10.1109/TITS.2024.3416508
16	KUMAR A, HARSHA S P A systematic literature review of defect detection in railways using machine vision-based inspection methods[J]. International Journal of Transportation Science and Technology, 2025, 18: 207- 226 doi: 10.1016/j.ijtst.2024.06.006
17	QIU Y, LIU H, LIU J, et al Region and edge-aware network for rail surface defect segmentation[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 3522213 doi: 10.1109/tim.2024.3406815
18	HUSSAIN M YOLO-v1 to YOLO-v8, the rise of YOLO and its complementary nature toward digital manufacturing and industrial defect detection[J]. Machines, 2023, 11 (7): 677 doi: 10.3390/machines11070677
19	WANG Q, WU B, ZHU P, et al. ECA-net: efficient channel attention for deep convolutional neural networks [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 11531–11539.
20	WU Y, QIANG F, ZHOU W, et al PFCNet: enhancing rail surface defect detection with pixel-aware frequency conversion networks[J]. IEEE Signal Processing Letters, 2025, 32: 606- 610 doi: 10.1109/LSP.2025.3525855
21	LIU Y, XIAO H, XU J, et al A rail surface defect detection method based on pyramid feature and lightweight convolutional neural network[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 5009510 doi: 10.1109/tim.2022.3165287
22	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 936–944.
23	LI X, XU X, YANG H A road crack detection model integrating GLMANet and EFPN[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25 (11): 18211- 18223 doi: 10.1109/TITS.2024.3432995
24	YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions [EB/OL]. (2016−04−30)[2025−06−02]. https://arxiv.org/pdf/1511.07122.
25	张振利, 胡新凯, 李凡, 等基于CNN和Efficient Transformer的多尺度遥感图像语义分割算法[J]. 浙江大学学报: 工学版, 2025, 59 (4): 778- 786 ZHANG Zhenli, HU Xinkai, LI Fan, et al Semantic segmentation algorithm for multiscale remote sensing images based on CNN and Efficient Transformer[J]. Journal of Zhejiang University: Engineering Science, 2025, 59 (4): 778- 786
26	WANG T, ZHANG Z, TSUI K L A hierarchical transfer-generative framework for automating multianalytical tasks in rail surface defect inspection[J]. IEEE Internet of Things Journal, 2024, 11 (12): 21513- 21526 doi: 10.1109/JIOT.2024.3374751
27	BONO F M. Railway computer vision model [DB/OL]. [2025−06−02]. https://universe.roboflow.com/francesco-morgan-bono-cntpn/railway-yyihi.
28	ZHAO Y, LV W, XU S, et al. DETRs beat YOLOs on real-time object detection [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024: 16965–16974.

[1]	李珺,丁彬彬,史维娟,杨琳. 基于多尺度特征聚合的航拍图像检测算法[J]. 浙江大学学报(工学版), 2026, 60(9): 1890-1900.
[2]	李国燕,于威,梅玉鹏,张明辉,王新强. 全局局部特征融合的遥感图像建筑物提取[J]. 浙江大学学报(工学版), 2026, 60(5): 1100-1108.
[3]	宋耀莲,彭驰,唐菁敏,赵宣植,虞贵财. 基于融合注意力机制的光学遥感图像小目标检测算法[J]. 浙江大学学报(工学版), 2026, 60(4): 763-771.
[4]	李国燕,李鹏辉,刘榕,梅玉鹏,张明辉. 融合多尺度分辨率和带状特征的遥感道路提取[J]. 浙江大学学报(工学版), 2026, 60(3): 585-593.
[5]	孟昱煜,孔垂乐,火久元,武泽宇. 重构YOLOv11的无人机小目标检测算法[J]. 浙江大学学报(工学版), 2026, 60(2): 303-312.
[6]	刘春娟,包易航,吴小所,闫浩文,朱铁栓,赵浩然. 基于细节增强和跨尺度注意力融合的PCB缺陷检测算法[J]. 浙江大学学报(工学版), 2026, 60(11): 2431-2441.
[7]	刘宇轩,苏恩祈,井佩光,孙宏博,苏育挺. 基于多任务对比约束与可控缺陷生成的异常检测算法[J]. 浙江大学学报(工学版), 2026, 60(11): 2442-2452.
[8]	张建刚,李肖,冯丹丹. 基于动态核感知的无人机视角路面病害检测方法[J]. 浙江大学学报(工学版), 2026, 60(10): 2141-2152.
[9]	翟亚红,陈雅玲,徐龙艳,龚玉. 改进YOLOv8s的轻量级无人机航拍小目标检测算法[J]. 浙江大学学报(工学版), 2025, 59(8): 1708-1717.
[10]	王浚银,文斌,沈艳军,张俊,王子豪. 基于改进YOLOv7-tiny的铝型材表面缺陷检测方法[J]. 浙江大学学报(工学版), 2025, 59(3): 523-534.
[11]	刘登峰,郭文静,陈世海. 基于内容引导注意力的车道线检测网络[J]. 浙江大学学报(工学版), 2025, 59(3): 451-459.
[12]	朵琳,殷瑜,段威,张芸,任勇. 基于改进YOLOv8的船舶目标检测算法[J]. 浙江大学学报(工学版), 2025, 59(11): 2379-2388.
[13]	刘欢,李云红,张蕾涛,郭越,苏雪平,朱耀麟,侯乐乐. 基于MA-ConvNext网络和分步关系知识蒸馏的苹果叶片病害识别[J]. 浙江大学学报(工学版), 2024, 58(9): 1757-1767.
[14]	刘毅,陈一丹,高琳,洪姣. 基于多尺度特征融合的轻量化道路提取模型[J]. 浙江大学学报(工学版), 2024, 58(5): 951-959.
[15]	曹寅,秦俊平,高彤,马千里,任家琪. 基于生成对抗网络的文本两阶段生成高质量图像方法[J]. 浙江大学学报(工学版), 2024, 58(4): 674-683.

Viewed

Full text

Abstract

Cited

Shared

Discussed